Subseasonal extremes of precipitation and active

Subseasonal extremes of precipitation and active-break cycles of the Indian summer monsoon
in a climate change scenario
By A. G. TURNER∗ , and J. M. SLINGO
Walker Institute, University of Reading, UK
(submitted 23 January 2008, revised 26 September 2008)
Summary
Changes to the behaviour of subseasonal precipitation extremes and active-break cycles of the
Indian summer monsoon are assessed in this study using pre-industrial and 2 × CO2 integrations
of the HadCM3 coupled GCM, which is able to reasonably simulate the monsoon seasonal cycle.
At 2 × CO2 , mean summer rainfall increases slightly, especially over central and northern India.
The mean intensity of daily precipitation during the monsoon is found to increase consistent with
fewer wet days, and there are increases to heavy rain events beyond changes in the mean alone.
The chance of reaching particular thresholds of heavy rainfall is found to approximately double
over northern India, increasing the likelihood of damaging floods on a seasonal basis. The local
distribution of such projections is uncertain, however, given the large spread in mean monsoon
rainfall change and associated extremes amongst even the most recent coupled climate models. The
measured increase of the heaviest precipitation events over India is found to be broadly in-line with
the degree of atmospheric warming and associated increases in specific humidity, lending a degree
of predictability to changes in rainfall extremes. Active-break cycles of the Indian summer monsoon,
important particularly due to their effect on agricultural output, are shown to be reasonably represented
in HadCM3, in particular with some degree of northward propagation. We note an intensification of
both active and break events, particularly when measured against the annual cycle, although there is
no suggestion of any change to the duration or likelihood of monsoon breaks.
This is a preprint of an article published in the Q. J. R. Meteorol. Soc. 135, 640, pgs 549-567.
http://dx.doi.org/doi:10.1002/qj.401
http://onlinelibrary.wiley.com/doi/10.1002/qj.401/abstract
c Royal Meteorological Society 2009
Copyright Keywords: Asian summer monsoon; global warming; rainfall extremes; intraseasonal oscillation; active-break events
1. Introduction
The Indian summer monsoon acts to tie the lives of millions of people across South Asia
to the seasonal cycle, through their dependence on its rainfall for agriculture and, increasingly,
industrial development. Climate change studies focusing on the monsoon have generally considered
how robust the monsoon will be in the changing global climate. However, variations in the monsoon
on subseasonal timescales are arguably of more importance to the local population. Short periods of
heavy rainfall can cause floods with devastating consequences. In late July 2005 for example, Mumbai
(Bombay) received nearly one metre of rainfall (Lal et al., 2006), leading to extensive damage to
infrastructure, disease, and loss of life. On the other hand, the extended monsoon break of July 2002
contributed to a country-wide drought, with All-India rainfall for the summer reaching only 81% of
its long period average (Annamalai et al., 2007). This led to a reduction in both agricultural output
and economic growth.
In describing the impact of increased greenhouse gas forcing experiments on the Asian summer
monsoon, Turner et al. (2007) cited several studies which noted that the monsoon remained generally
robust. At 2 × CO2 , increases in mean Indian monsoon precipitation of 0.5–2mm/day were noted
over India and the Bay of Bengal. A slight northward displacement of the south-westerly jet was
also noted by Turner et al. (2007) using HadCM3, together with higher interannual variability.
Both studies attributed increases in monsoon precipitation to the warmer Indian Ocean providing
additional moisture. There was also a limited increase to the south-westerly flow over the northern
Corresponding author: Walker Institute for Climate System Research, Department of Meteorology, University of Reading, PO
Box 243, Earley Gate, Reading, RG6 6BB, UK. email: [email protected]
∗
1
2
Arabian Sea in Turner et al. (2007), attributable to the increased land-sea temperature contrast at
2 × CO2 . While Hu et al. (2000) noted increased precipitation over only peninsular India, Ashrit
et al. (2003) saw increases over the drier regions of the northwest and far south, and others noted
overall intensification of the mean pattern (Kitoh et al., 1997; May, 2004). Sometimes such increases
in monsoon precipitation were even in regions of weaker monsoon flow regimes (Kitoh et al., 1997;
Meehl et al., 2000; Ashrit et al., 2003), highlighting the model-dependent response of the monsoon to
increased greenhouse gas forcing. In a study of the CMIP3 archive (Coupled Model Intercomparison
Project data archived at the Program for Climate Model Diagnosis and Intercomparison; PCMDI),
Annamalai et al. (2007) noted that of the six models which reasonably simulated the present day
seasonal cycle of precipitation in the Indian region, all featured increased monsoon rainfall at
doubled CO2 . The Intergovernmental Panel for Climate Change (IPCC) Fourth Assessment report
also suggests future increases in precipitation of the Asian summer monsoon (Meehl et al., 2007),
although the multi-model ensemble mean response is smaller than the inter-model standard deviation.
Even with an apparently robust monsoon under increased greenhouse gas forcing, changes to the
intensity of extreme events and the duration of drought spells could have much more devastating consequences, note Allen and Ingram (2002). While these authors suggest that global mean precipitation
increases are constrained by the balance between radiative and latent heating, they propose that the
heaviest rainfall events occur when all moisture in the column precipitates out.
The increased magnitude of these extreme events should be predictable according to the available
moisture (i.e., Clausius-Clapeyron) at a rate approximating 6.5%K−1 in their study. Trenberth et al.
(2003) subsequently noted that in warmer climates where atmospheric moisture is rising more quickly
than the global mean precipitation amount, increases in rainfall intensity must be offset by decreases
in frequency and/or by decreased light to moderate amounts of rainfall.
To help understand how these characteristics may change in the future, we first review studies
based on the recent observed record. Sen Roy and Balling Jr. (2004) studied 1910–2000 raingauge
data from 129 stations spread across India, and considered seven indices measuring extreme heavy
rainfall events (including largest daily rainfall totals and upper precipitation percentiles), giving a
total of 903 timeseries. Of these, 61% showed an upward trend, 114 being significant. 61 timeseries
had a significant downward trend. Increases were generally found spanning northwest Himalaya to
peninsular India, whilst decreases were noted over the East Gangetic Plains. Goswami et al. (2006)
puzzled over the apparent stability of precipitation over 1951–2000 from the India Meteorological
Department (IMD) gridded dataset, given the rise in global surface temperatures over the same
period. Focusing on the central Indian region, they noted significant rising trends in the frequency
and magnitude of extreme rainfall events, and conversely a decreasing trend in the frequency of
moderate events. Such compensatory changes led to little climate change signal in seasonal mean
rainfall. Extrapolating such trends with further increases in greenhouse gas forcing would lead to
increased flood hazards over central India in the future, conclude Goswami et al. (2006).
When periods of heavy rainfall last for several days, they may be thought of as active events of
the monsoon. Break events form the opposing phase of this cycle of intraseasonal oscillation and can
have severe impacts on agriculture, particularly if they extend for a week or more. Krishnamurthy and
Shukla (2000) give a thorough account of the phenomenology of active-break cycles over the Indian
subcontinent based on gridded rainfall data prepared from 3700 stations. Highlighting the importance
of understanding interactions between seasonal mean rainfall and active-break events, they find that
heavy drought years feature negative rainfall anomalies covering nearly all of India and persisting for
the entire season. Classifying active and break events according to All-India rainfall, Krishnamurthy
and Shukla (2000) note active (break) phases to feature above (below) normal precipitation over
central India and below (above) normal over northern India towards the Himalyan foothills, and
southern India. When atmospheric fields are studied over larger scales however, more features emerge.
Annamalai and Slingo (2001) composited outgoing longwave radiation (OLR) anomalies from
ECMWF reanalysis data 1979-95 based on All-India rainfall data. They noted peak power in two time
bands (10–20 and 40–60 day), explaining 25% and 66% of total intraseasonal variability respectively.
During active events, the authors found enhanced convection over the Indian subcontinent, extending
into the Bay of Bengal and south-eastward down to the Maritime Continent and equatorial west
Pacific. Reduced convection to the south of India and in the East China Sea completes a quadrupole
pattern of coherent anomalies. The 10–20 day mode was noted to be on a smaller scale however.
Goswami and Xavier (2005) also noticed the difference in zonal scale of the different modes, being
quite narrow at 10–20 days and much broader on longer timescales (30–90 days) in their study of
3
NCEP-NCAR zonal wind reanalysis data at 850hPa. They also highlight propagation directions,
noting the 10–20 day mode to feature westward propagation at Indian latitudes, whilst the 30–90 day
mode depicts northward propagation. In their principal oscillation pattern (POP) analysis, Annamalai
and Slingo (2001) also note the northward propagation of the long-period mode, highlighting its
origins over the equatorial Indian Ocean. Thus active-break cycles can be considered as a component
of boreal summer intraseasonal oscillations (BSISO Kemball-Cook and Wang, 2001; Fu et al., 2003,
and many others) involving complex interactions between northward and westward propagations,
often accompanied by eastward progressions in the equatorial Indian Ocean which are of similar
character to the Madden-Julian Oscillation (MJO).
In a more recent study based on 70 years of high resolution gridded rainfall data from the IMD,
Krishnamurthy and Shukla (2007) instead envisage the propagation of active-break anomalies over
India as southwest to northeast transitions of the monsoon trough from its preferred position over
central India to the Himalayan foothills. Then, on shorter timescales, westward movement of monsoon
depressions occurs within the trough itself. Providing a more detailed description than their earlier
study, they describe active events beginning over the Western Ghats and east-central India, intensifying
and expanding to cover the whole of central and parts of north India. Meanwhile negative precipitation
anomalies reach towards the Himalaya and exist separately in the south eastern states. The final phase
of the event sees positive anomalies moving toward the Himalayan foothills while the south peninsula
is covered by reduced precipitation. In a subsequent study, Krishnamurthy and Shukla (2008) describe
a direct relationship between the variability of convection over the broad Indian domain and rainfall
over India itself.
In the observed record, disagreement in changes to active-break cycles is noted between several
studies. Joseph and Simon (2005) describe the duration of break periods as increasing by 30% over
the 1950–2002 period using 850hPa zonal flow from the NCEP-NCAR reanalysis. However here they
define duration as the number of break days in the season, neglecting their consecutive behaviour.
Using raingauge data provided by the Indian Institute for Tropical Meteorology (IITM) and defining
break (active) thresholds as p < 8mm/day (p > 12mm/day ), they note changes in number of +45%
and −78% respectively. Rajeevan et al. (2006) however, in their study of 1◦ gridded data from
1951–2003, showed no significant trends in the number of active and break days in the season. This
perhaps relates to their definition of events relative to standardised rainfall anomalies. In one of the
few modelling studies, Mandke et al. (2007) study 10 coupled GCMs in the CMIP3 database, but
confine their study to six, based on their present day simulation of the seasonal cycle. They define
events based on precipitation over central and northwest India, and demonstrate that the response to
climate change is uncertain among the models and even between 1pctto2x and 1pctto4x experiments
for a given model. Some models measured showed increases in the number of active days whilst
other showed decreases, however the intensity of break events and their spatial extent are shown
to increase in all (some) of the models tested at 4 × CO2 (2 × CO2 ). Relatively few studies have
attempted to measure changes to active-break events in greenhouse gas forced GCM experiments,
perhaps relating to doubts over the ability of GCMs to simulate the seasonal cycle of precipitation and
features on regional scale. This paper aims to consider the impact of increased greenhouse gas forcing
on intraseasonal monsoon variability in the Hadley Centre coupled GCM (HadCM3), characterized
by both precipitation extremes on subseasonal timescales and active-break cycles together. This will
provide a framework from which future, better resolved models, may be examined. Section 2 describes
the datasets and model integrations studied, together with the choice of index used to define active and
break events. In section 3, the modelled monsoon climate, its seasonal cycle and BSISO behaviour
are compared with observed data. Section 4 shows how climate change manifests itself onto extremes
of monsoon precipitation and the effect of increased greenhouse gas forcing on active-break cycles of
the monsoon. Conclusions are drawn in section 5.
2. Method and data
(a)
Model integrations
The UK Met Office Unified Model HadCM3 is used in this study, comprising fully coupled
atmosphere and ocean components (Pope et al., 2000; Gordon et al., 2000) which exchange information once per day. The model features a stable mean state over several centuries, requiring no input of
artificial heat flux adjustments at the ocean surface to combat climate drift (Johns et al., 2003). While
4
a previous study (Turner et al., 2005) found that use of an annual cycle of equatorial Indo-Pacific heat
flux adjustments could improve the behaviour of monsoon interannual variability and its interaction
with ENSO in this model, no improvements were found in monsoon intraseasonal variability and so
they have not been employed here. The atmosphere is solved on a 3.75◦ longitude by 2.5◦ latitude
grid, and as in previous studies by the authors, the atmospheric component has been run with 30
levels in the vertical rather than the standard 19 (giving higher resolution in the mid-troposphere), due
to better representation of intraseasonal tropical convection such as the MJO (Inness et al., 2001) and
a more accurate precipitation response to high SST forcing (e.g. El Niño, Spencer and Slingo, 2003).
The ocean resolution is 1.25◦ in the horizontal with 20 vertical levels. HadCM3 provides an accurate
simulation of the spatial structure of the monsoon and its seasonality (Martin et al., 2000) as well as
good representation of the monsoon trough (Johns et al., 2003), important for simulating variations
on subseasonal timescales. In addition, it was used extensively for climate change studies cited in the
IPCC Third and Fourth Assessment Reports. Indeed of 18 models from the CMIP3 database which
Annamalai et al. (2007) tested, HadCM3 was one of only 6 judged to reasonably simulate the present
day seasonal cycle of the monsoon.
This study uses results from a control run of HadCM3 with pre-industrial greenhouse gas forcing
(approximately 290p.p.m., hereafter 1 × CO2 ), initialised from a previous L19 integration. An initial
period of 10 years was discarded to allow for any spin-up to the increased vertical resolution, before
a further 100-year integration was performed. The climate change experiment was initialised from
an existing 150-year 2 × CO2 run, prior to which carbon dioxide had been ramped up at a rate of
1%/year for 70 years, in accordance with IPCC guidelines. The first 10 years of the new experiment
were again discarded as a cautious adaptation to L30, before a 100 year integration at 2 × CO2
(580p.p.m.) was performed. No sulphate forcing was employed.
(b)
Observed data
In order to briefly validate the model used, three datasets have been obtained for comparison. The
Climate Data Center Merged Analysis of Precipitation (CMAP, Xie and Arkin, 1997) has been used
to assess mean precipitation. This consists of a combination of satellite and gauge data over the period
1979–2004 with monthly output. An entirely observational dataset has also been used. The India
Meteorological Department (IMD) have developed a 1◦ -gridded dataset based on spatial averaging of
station data. The original dataset (Rajeevan et al., 2006) took information based on 1803 stations for
the period 1951–2003, with each station available for at least 90% of days over that period. However
in these data the station density over the northern plains was low. Here, a new version of the dataset
is used (M. Rajeevan, personal communication, 2007), comprising 2140 stations (mainly increasing
density over the northern plains) and extension to 2004. The advantage of these data is that they are
daily and thus well suited for comparison with the model seasonal cycle. In addition, to study daily
characteristics over the broad Indian Ocean domain, the Global Precipitation Climatology Project 1◦ gridded daily dataset (Huffman et al., 2001; Adler et al., 2003) has been used. This covers only the
recent period 1997–2007.
(c)
Choice of active-break index
Many different indices have been used to generate composite evolutions of active-break events
during the Indian summer monsoon. Perhaps due to the poor availability of quality long-period
precipitation data, several studies based their indices on circulation obtained from reanalysis products.
Goswami and Mohan (2001) recognised the stronger monsoon trough during active periods and
inferred that westerlies south of the trough would strengthen during such events. They based their
index on 30–60-day filtered 850hPa zonal winds at 90◦ E, 15◦ N, although their results were not too
dependent on the exact point chosen. Webster et al. (1998) too chose a zonal wind-based index over the
north Bay of Bengal, and selected active (break) events outside absolute anomaly values of (−)3ms−1 .
Goswami and Mohan (2001) however chose events based on standard deviations from the mean.
The proliferation of satellite data means that good OLR datasets are available for use as proxies
for convection. Vecchi and Harrison (2002) devised a monsoon break index based on the difference
between normalized 7-day boxcar smoothed OLR anomalies between two regions: Indian (10–30◦ N,
65–85◦ E) and eastern equatorial Indian Ocean (10◦ S–5◦ N, 75–95◦ E). The 50-day smoothed anomaly
was also removed to reduce the effect of prevailing large-scale monsoon conditions on the resultant
composites. Their index naturally captures the dipole of convection between the oceanic TCZ region
5
and Indian latitudes, and the transition and northward movement of convection between them during
active-break cycles.
Precipitation, however, offers more insight into monsoon activity at the surface and its impact
on the local population. Some authors average precipitation over central India, representing the core
monsoon region (e.g. Mandke et al., 2007, uses 73–82◦ E, 18–28◦ N) due to the presence of the largest
anomalies on intraseasonal timescales over this region, while opposite signed anomalies lie to the
northeast and southeast. A similar region is chosen by Goswami et al. (2006) to represent central India
(74.5–86.5◦ E, 16.5–26.5◦ N). Krishnamurthy and Shukla (2000, 2007) however chose events based
on rainfall over the all-India region. Area-averaged precipitation over the Indian land surface has
its climatological seasonal cycle removed before being normalized by its standard deviation. Active
events exceed +1σ while break events fall below −1σ , each for 3 or more consecutive days (5 or more
in their 2007 paper). One may argue that the northward propagating nature of such events precludes
using data from the whole country, however the All-India index still captures the dominant signal
from the core monsoon region where seasonal mean rainfall and its anomalies are at their largest. In
any case, Krishnamurthy and Shukla (2007) still demonstrate northward propagation of precipitation
anomalies across India in the observed record. Additionally Annamalai and Slingo (2001) generated
OLR composites based on All-India rainfall, and showed that coherent convective structures exist on
a much broader scale than India, confirming that active-break events do not need to be defined in such
small regions.
For model data available at the horizontal resolution of HadCM3, using data from all-India has the
benefit that more data points are considered compared to a smaller region. In this study, the gridpoints
were chosen after Gadgil and Sajani (1998), representing the 27 Indian land-surface points within 10–
27.5◦ N, 71.25–86.25◦ E. Thus model data are averaged over the chosen gridpoints and the seasonal
cycle is removed, leaving the anomalous rainfall at year y and day d as p′ :
p′ (y, d) = p(y, d) −
100
1 X
p(y, d).
100
(1)
y=1
These data are then normalized by the cycle of standard deviation through the season. Composite
active and break cycles are chosen where the index exceeds ±1σ for five or more days.
3. Model simulation of monsoon rainfall
(a)
Mean, seasonal cycle and probability distribution
Before measuring if subseasonal extremes of monsoon variability and active-break cycles will
change in the future, we perform a short validation of the simulated monsoon climate in HadCM3, its
seasonal cycle, and how these may change in the future.
Figure 1 shows the seasonal mean (June to September) monsoon rainfall in HadCM3, compared
with satellite derived data (CMAP, 1979-2004) and 1◦ -gridded IMD gauge data (1951–2004). The
general precipitation climate of the northern Indian Ocean region in HadCM3 is good in comparison
with CMAP, although the maximum in the eastern equatorial Indian Ocean extends too far to the
south-east, and is too dry at 60◦ E. The simulated distribution over India and the Bay of Bengal
is reasonable. Differences between the model and CMAP are no larger than those between CMAP
and IMD. The IMD dataset benefits particularly from its high resolution, allowing intense rainfall
upstream of the Western Ghats to be well captured, together with the downstream rain shadow. The
low-level monsoon wind climatology of HadCM3 is good, if a little strong, as already described in
Turner et al. (2005). Fig. 1d shows the climate change response of summer monsoon precipitation
to CO2 doubling. Turner et al. (2007) have already noted these significant precipitation increases
across north India, the head of the Bay of Bengal and also the Indian Ocean off southern India, a
spatial distribution well resembling that in the study of Meehl and Arblaster (2003) using the Parallel
Climate Model (PCM). In HadCM3, these increases amount to around 5% when summed over the
broad South Asia region (Turner et al., 2007), and are made up largely by increases in convective
rainfall (not shown). This is in agreement with local thermodynamic influences on increases in tropical
monsoon precipitation in the multi-model ensemble of future climate integrations in the CMIP3
database assessed by Sun et al. (2007).
6
(a)
(b)
5
5
7.5
15N
5
10
7.5
0.5
EQ
1
15
3
10
10
30N
2
5
3
7.5
1
5
10
7.5
5
10
5
10
15
10
10
3
2
15S
3
5
15N
2
1
3
1
2
2
30E
60E
0
0.5
30N
90E
1
3
7.5
2
120E
1
10
20
5
15
15
5
5
3
5
1
15N
3
1
0.5
7.5
10
5
10
1
EQ
5
10
0.5
0.5
2
0.5
2
1
1
-0.5
-0.5
-0.5
-2
15S
2
3
4
5
EQ
15S
2
(d)
3
1
5
5
10
30S
30E
30N
7.5
1
3
0.5
(c)
15N
90E
15
5
3
-0.5
-1
0.5
-0.5
2
60E
0.5
90E
120E
30S
30E
60E
90E
120E
Figure 1. Seasonal mean summer (JJAS) precipitation climatology in (a) CMAP, (b) IMD gridded data, (c) HadCM3 1 × CO2
and (d) HadCM3 2 × CO2 minus 1 × CO2 . Units are mm/day. Left colour-bar is for the mean climate panels (shaded only when
exceeds 5mm/day), right colour-bar is for climate difference panel (positive contours solid, negative contours dashed).
Figure 2. Mean seasonal cycle for days since June 1 in the IMD gridded dataset (dotted), HadCM3 1 × CO2 (solid) and HadCM3
2 × CO2 (dashed). Also shown for comparison are the GPCP daily data interpolated to the HadCM3 grid (grey dot-dashed). Units
are mm/day. Right-hand axis and top section show the significance level of the difference between 2 × CO2 and 1 × CO2 curves
using a student t-test, values below 90% are omitted. Curves are smoothed using a five day running mean for clarity.
To determine when during the season these changes are occurring and for validation, the
climatological seasonal cycle is presented in Fig. 2. For the model, this is determined by areaaveraging precipitation over the Indian land surface (as defined in section 2.3). For the spatial mean
of IMD data we first remove the detached northeastern states (Assam etc.) as these are not included
in the HadCM3 mean. HadCM3 does quite a reasonable job of simulating the seasonal cycle of the
monsoon as mentioned in the introduction. The two main discrepancies are the late onset in HadCM3
(around two weeks) and deficient rainfall following the peak in mid-July through to the withdrawal.
As an alternative, Fig. 2 also shows a comparison against that calculated from the satellite-derived
GPCP daily data 1997–2007, which has been interpolated to the model grid such that exactly the same
points are used to represent Indian rainfall. This comparison reveals the weak onset and withdrawal
to be robust errors in HadCM3, however rainfall during the peak period is lower than HadCM3,
with wide variability in the GPCP curve reflecting the short data-period of this record. It is worth
noting that the monsoon seasonal cycle of HadCM3 compares very well against 5 other models of
the CMIP3 database, chosen due to their good representation of the spatial mean during the summer
season (Annamalai et al., 2007, their Fig. 1h). The overall weak bias of the seasonal cycle reflects
subtleties in the spatial distribution of monsoon rainfall, as Turner et al. (2007) found HadCM3 to have
a negative precipitation bias against observations over India itself, yet a positive bias when measured
7
(a)
(b)
Figure 3. Probability distribution functions of (a) area-averaged Indian rainfall and (b) Indian rainfall with no spatial averaging in
the IMD gridded dataset (dotted), HadCM3 1 × CO2 (solid) and HadCM3 2 × CO2 (dashed). In (b), grey curve shows IMD data
when downgraded to the N48 grid of HadCM3. Bins are 0.5mm/day wide.
over the broader South Asian monsoon region. Biases during the onset may inhibit intraseasonal
activity during the early part of the season (Kemball-Cook and Wang, 2001). At 2 × CO2 , increases
in daily precipitation generally occur from mid-July to late September, significant above the 95%
level.
To look at the distribution of rainfall events in HadCM3, probability distribution functions
(hereafter pdfs) of daily data are shown in Fig. 3. This comprises two methods: (a) a combination of all
data as an area average and (b) each gridpoint. The area-average in Fig. 3a is performed over 27 model
grid-points of the Indian land-surface (and all data in the IMD grid), and precipitation is divided into
0.5mm bins. HadCM3 1 × CO2 simulates the gross features of the rainfall distribution, with the most
likely daily precipitation total being around 8mm in model and observations. However, at low rainfall
amounts the curves reveal a high degree of spatial coherence in HadCM3. That is, when it drizzles in
HadCM3 it is often doing so over the whole subcontinent. IMD is based on station data and provides
a better representation of small-scale convective events in response to India’s varied orography, versus
smoother orography and convection in HadCM3. Precipitation is thus more inhomogeneous over India
in nature, reducing the count to zero at low values.
A simple skill score to test the common area between two pdf curves was devised by Perkins
et al. (2007), and is reproduced here:
n
X
min(fobs , fmod ),
(2)
Sscore =
1
where the score is summed over all n bins, and fobs , fmod are the precipitation frequencies for a given
bin in the observations and model respectively. Values of Sscore approaching 1 indicate a high skill
in simulating the distribution of precipitation. Comparing the IMD and HadCM3 1 × CO2 curves in
Fig. 3a, we have Sscore = 0.77, which is quite reasonable (and above the cutoff which Perkins et al.,
2007, use to discard poorly-performing models in their study of Australian rainfall).
Alternatively, Fig. 3b shows rainfall pdfs for all grid-points with no spatial averaging and on a
log-frequency scale, allowing for improved interpretation at high rain rates. Here the 1 × CO2 model
curve is similar to that of the IMD gridded data, including a very high frequency of drizzle. Sun et al.
(2006) noted too frequent precipitation at low intensity to be a common problem amongst GCMs
(after Dai and Trenberth, 2004). This is in part related to the gridding, which increases precipitation
frequency at the expense of reduced intensity. Hennessy et al. (1997) pointed out that one would
expect GCMs to underestimate the frequency of heavy convective precipitation (as in the monsoon),
because this occurs on sub-grid scales and is thus being spread over a larger area in HadCM3. Sun
et al. (2006), however, demonstrated that the difference between grids of 1◦ (IMD) and 3◦ (HadCM3,
approximately) were small compared to differences between models, or between gridded and station
data. It is clear from Fig. 3b that, when looking at grid-points separately, HadCM3 is overestimating
moderate rainfall at the expense of events in the upper tail. To test the effect of the grid size itself,
the grey curve in Fig. 3b represents the IMD data downgraded to the HadCM3 grid. This is achieved
by extrapolating IMD data over missing ocean regions (the Arabian Sea and Bay of Bengal) before
bi-linearly interpolating to the 3.75◦ × 2.5◦ spacing of HadCM3. The same 27 Indian land-surface
grid points are selected as in HadCM3, reducing the number of points from 357 to 27 and causing
increased noise in the tail. There is clear evidence of a reduction in precipitation intensity in the upper
tail due to data gridding alone.
8
(a)
(b)
0.09
0.08
0.08
0.01
0.00
512.0
1024.0
1
2528.0
6.0
.032.0
8.0
512
12
.0
0.02
64
8.0
21..0
0
0.03
4.0
256.0
0.01
0.05
0.04
0.03
0.02
0.06
4.
0
0.04
2.0
4.0
omega
0.05
2.0
omega
0.07
0.06
4.0
0.07
1.0
0.10
0.09
16.0
0.10
4.0
Figure 4. Normalized power spectra of daily rainfall anomalies averaged over the Indian region in (a) IMD gridded data, excluding
the far northeast states and (b) HadCM3 1 × CO2 . Spectra are calculated for each JJAS season separately before being smoothed
using a Tukey window. The theoretical spectra of an AR(1) process (first-order autoregressive) are used to distinguish peaks from
red noise.
0 0 .0
8.16.
4
32.0 6
0.00
-12 -10 -8 -6 -4 -2 0
2
4
6
meridional wavenumber
8 10 12
-12 -10 -8 -6 -4 -2 0
2
4
6
8 10 12
meridional wavenumber
Figure 5. Space-time spectra on
2–128 day filtered precipitation anomalies in (left) GPCP and right (HadCM3) over the domain
0.10
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
70–100◦ E, 30◦ N–30◦ S. Contours are powers of two, those above 16 are shaded.
Both parts of Fig. 3 suggest that at doubled CO2 , there is a decreased probability of moderate
rainfall, compensated for by increases in heavy rainfall events, particularly in the upper tail. This is
consistent with the study of Semenov and Bengtsson (2002) who noted increased heavy precipitation
at the expense of lower intensities in their IS92a scenario of the ECHAM4/OPYC3 model combination. Similarly, Hennessy et al. (1997) noted a general decreased probability of moderate rainfall
against increased frequencies at low and heavy levels. The largest increases in their study of two
coarse-resolution slab-ocean models were for very heavy rainfall.
(b)
Evidence for BSISO
In order to later examine changes to active-break cycles of the monsoon in HadCM3, here we
briefly validate intraseasonal behaviour during the boreal summer (BSISO) and discuss some of the
inherent biases in the model. Figure 4 shows normalized power spectra for the daily all-India rainfall
spatially-averaged over the IMD grid or Indian land surface gridpoints in HadCM3 (see section 2.3).
The spectra are calculated over each JJAS season separately before an average is taken. In HadCM3,
major significant peaks are evident at around 10–20 days and 40 days. These are broadly consistent
with the periods described in studies cited in the introduction. In the IMD dataset, somewhat more
‘weather noise’ is evident at short periods (< 10 days), however a significant peak is noted between
30 and 40 days.
To test propagation characteristics, we calculate space-time spectra over the Indian domain during
the May to October season, in order to also capture any intraseasonal variability associated with
the monsoon onset and withdrawal. Data is first filtered to intraseasonal (2–128 day) timescales
using a Lanczos filter (Duchon, 1979) before being zonally averaged. Then fast Fourier transforms
convert rainfall data in the (y, t) domain to wavenumber-frequency space (ly , ω). The space-time
spectra shown in Fig. 5 are given by the square of the absolute value of the Fourier coefficients,
averaged over each year in the datasets. As shown by the figure, northward propagation is primarily
at the gravest meridional wavenumber (Lin et al., 2008; Fu et al., 2003) in GPCP observations. In
HadCM3, northward power is not quite as distinct as in observations and concentrated at lower
wavenumbers / higher frequencies, suggesting slower phase speeds. HadCM3 also lacks obvious
variance in southward modes (negative meridional wavenumbers).
Finally, to demonstrate BSISO propagation more clearly we use lag-correlations of 30–60 day
bandpass-filtered precipitation during the extended boreal summer (MJJASO) in the Indian domain
-15
-10
-5
0
5
10
15
20
latitude (degrees)
-15
-10
-5
0
5
10
20
latitude (degrees)
0.5 0.3
0.1
0.5
0.1
0.1
20
0
60
120
180
240
300
360
longitude
(a)
-5
0
5
10
15
20
-20
-15
-10
-0.1
0.3
0.5
15
5
15
20
-20
0
10
15
20
-20
5
-5
-0.1
0.1
-0.1
-0.3
20
0
10
15
.3
-0
-0.1
5
10
-0 0
.1 .1
-0
.5
15
0
-10
-5
0.3
10
1
0.
-0.3
5
-5
-0.1
0
-15
-10
0.5
lag (days)
0.7
-20
-15
0.3
-5
-20
0.3
lag (days)
-10
.1
0.1
0.5
-0
0.1
-10
-15
3
-0.
-0
-0 .5
.3
.1
-0
1
lag (days)
-20
0.1
0.3
0.
-15
lag (days)
-20
-0.1
9
0
60
120
180
240
300
360
latitude
(b)
-10
-5
0
5
10
15
20
-20
-15
Figure 6. Lag-correlations of 30–60 day filtered rainfall anomalies during May to October showing northward and eastward
propagations in the Indian monsoon region after Lin et al. (2008): (a) zonally averaged over 70–100◦ E against itself near
12.5◦ N, 85◦ E, (b) meridionally averaged over 5–25◦ N against itself near to 15◦ N, 95◦ E. Positive contours solid, negative contours
dashed and shaded. In (a), the thick dashed lines represent approximate phase velocities of 0.8, 1.8 and 2.8ms−1 . Shown are GPCP
observations (left) and HadCM3 (right).
against some reference point. Figures 6a,b depict northward and eastward propagation respectively.
We choose identical domains and reference points to Lin et al. (2008) so that a direct comparison
can be made with the multi-model data used in their study. Figure 6a shows that HadCM3 is clearly
capable of some northward propagation in the Indian domain, to a degree comparable with many of
the other models in the CMIP3 database (see Lin et al., 2008, Fig. 16) and GPCP observations. Also
shown in the figure are lines indicating the same phase speeds as in Lin et al. (2008). Although
they found northward propagations in GPCP to be around 1.8ms−1 , here the propagation seems
slightly slower, perhaps due to the longer dataset used here. In HadCM3, northward propagations are
slightly slower than 0.8ms−1 . HadCM3 also shows an absence of southward propagation from around
5◦ S, although this should not affect intraseasonal activity over India itself. Both these findings are
consistent with Fig. 5. To show that northward propagation is not simply a manifestation of eastward
motion of a northwest-southeast tilted rainband (as implied in Fu et al., 2003), Fig. 6b shows eastward
propagations at Indian latitudes. These are very weak in HadCM3 when compared to GPCP data, and
consistent with Lin et al. (2008) who showed little correspondence between northward and eastward
BSISO modes in most CMIP3 models. Kemball-Cook and Wang (2001) suggested the importance of
air-sea interactions over the Indian Ocean for propagations associated with the BSISO and the poor
representation of these interactions in coupled models, together with weakness of the seasonal cycle
during the monsoon onset (Fig. 2), may be the cause of such biases in HadCM3.
4. Results
(a)
Changes to precipitation extremes
Changes to precipitation extremes can manifest themselves both in the intensity, duration, and
likelihood of events. This section considers the impact of increased greenhouse gas forcing on changes
in precipitation event characteristics.
A first measure of the reliability of precipitation during the monsoon season can be given by the
number of wet days, or the proportion of the season in which rain occurs. In the IMD data, roughly
60% of days during the season JJAS are wet over central India, rising above 70% over the Western
Ghats and 80% in the northeast, while falling to 20% in the dry north-western states, and around
30% in the southeast (not shown). Using daily Indian station data from 1986 to 1989, Stephenson
et al. (1999) saw a range of 40–60% wet days over peninsular India and fewer over the northwest
deserts. The occurrence of wet days through the monsoon season is shown for HadCM3 in Fig. 7(a).
To avoid computational and truncation errors manifest in the output of model precipitation, a low
precipitation cut-off of 0.1mm/day is used to define a wet day, below which days are defined as dry
(after Stephenson et al., 1999; Semenov and Bengtsson, 2002). Under control conditions, HadCM3
reasonably simulates the spatial variation in occurrence of wet days, between 60–70% over much of
central India and rising above 70% in the northeast and over the Western Ghats. In the southeast, the
proportion of wet days during the season reaches similar levels to the observed dataset, although the
signal is displaced somewhat into the Bay of Bengal. This likely reflects the coarse model resolution
and poor simulation of the rain shadow east of the Western Ghats.
10
(a)
(b)
Figure 7. Percentage number of wet days per season in (a) HadCM3 1 × CO2 and (b) HadCM3 2 × CO2 minus 1 × CO2 .
Rainfall is defined as occurring where precipitation exceeds 0.1mm/day on a given day.
30N
0.5
1
30N
7.5
15N
1
15
15
1
7.5
5
1
EQ
2
5
10
10
1
15S
5
30S
30E
5
1
0.5
90E
3
7.5
2
5
(a)
120E
1
-2
0.5
-1
-0.5
30S
30E
60E
15
10
0.5
-0.5
120E
90E
1
20
1 1
2
1
3
60E
0
1
0.5
0.5
-0.5
EQ
15S
0.5
3
15N
0.5
1
1
4
1
0.5
15
10 7.5
7.5
5
3
1
1
7.5
10
3
0.5
2
5
4
(b)
Figure 8. (a) Summer mean rainfall intensity in HadCM3 and (b) HadCM3 2 × CO2 minus 1 × CO2 . Rainfall is defined as
occurring where precipitation exceeds 0.1mm/day on a given day.
At 2 × CO2 there are more than 2% fewer wet days per season (shown in Fig. 7(b)), amounting
to an extra three days with no rainfall. This figure rises rapidly out into the southwest Bay of Bengal.
Semenov and Bengtsson (2002) also noted a clear reduction in the probability of wet days occurring
in the ECHAM4/OPYC3 coupled model between transient simulations of the 20th and 21st centuries.
The number of wet days can be related to the precipitation intensity, hence mean precipitation
intensity is calculated as the total seasonal precipitation divided by the number of wet days in the
season at each gridpoint. In observations, precipitation intensity is generally wetter for all areas of
India than the corresponding seasonal mean (not shown), but is qualitatively similar. This confirms
the discrete nature of rainfall events, many of which are very intense (Stephenson et al., 1999). Mean
precipitation intensity is shown for HadCM3 in Fig. 8a, and is seen to be qualitatively similar to (but
larger in magnitude than) the corresponding seasonal mean (Fig. 1c). The model features generally
reduced levels of precipitation intensity in comparison with the IMD observed data. This can be
partially attributed to the coarse model grid (each gridbox in HadCM3 represents more than 9 times
the area of a 1◦ box in the IMD data), which acts to smooth out heavy events (Sun et al., 2006).
The response of precipitation intensity to increased greenhouse gas forcing is large (see Fig. 8b),
especially over northern India and the north Bay of Bengal, reaching +2–4mm/day. This pattern is in
spatial agreement with the climate change response of mean precipitation (Fig. 1d), although intensity
increases to a greater degree, especially over northern India. Given that intensity is proportional to the
mean and varies inversely with the number of wet days, this finding is consistent with the decreased
number of wet days seen in Fig. 7(b). The only intensity decreases noted in our study occur over
south-east peninsular India, a region which features little mean change but more wet days, and in any
case one of the drier regions during the Indian summer monsoon.
To examine the spatial detail of heavy rainfall events on subseasonal timescales, the levels of
precipitation at heavy (defined as the 95th) and very heavy (99th) percentiles are shown in Fig. 9
11
16
24
4
30N
24
20N
40
16
EQ
60E
75E
90E
105E
EQ
60E
120E
(d)
30N
75E
24
20N
10N
105E
EQ
60E
120E
32
105E
120E
4
8
20N
70
10N
4
12
8
10N
24
2
90E
4
16
90E
4
75E
30N
32
40
24
16
75E
0
2
4
(f)
20N
60
8
4
90E
30N
32
32
2
8
10N
24
(d)
8
EQ
60E
16
4
2
20N
16
16
2
6
24
50
10N
24
16
30N
24
32
2
16
20N
10N
(c)
(b)
(a)
30N
105E
EQ
60E
120E
16
32
8
75E
50
24
40
24
16
90E
70
60
105E
4
EQ
60E
120E
80
75E
4
2
90E
105E
8
6
2
120E
12
10
14
Figure 9. Spatial patterns of extreme daily rainfall at the heavy (95th percentile) level: (a) HadCM3 1 × CO2 , (b) HadCM3
2 × CO2 , and (c) their difference. (d) to (f) show the same but for very heavy (99th percentile) levels of precipitation. Percentiles
are calculated on a season-by-season basis and the mean is taken. Absolute panels feature shading above 24mm/day, difference
panels are shaded above 2mm/day. Only those differences significant above the 95% level are shown, there are no significant
negative changes.
30N
(a)
30N
20N
20N
10N
10N
EQ
60E
75E
90E
105E
0
EQ
60E
120E
2
1
(b)
75E
90E
105E
120E
4
3
5
Figure 10. Percentage change in the likelihood of reaching 1 × CO2 precipitation levels: (a) heavy (95th percentile) and (b) very
heavy (99th percentile) as CO2 concentrations are doubled. A 5% (1%) increase in the chance of (very) heavy rainfall amounts to
a doubling, by definition.
for both climate scenarios. These levels of precipitation are exceeded on approximately 6 days and 1
day per season. To avoid the effects of interannual variability (and hence clustering of heavy rainfall
in a particular year), percentiles are calculated separately for each season before the mean is taken.
The differences in the levels of precipitation associated with these upper percentiles are shown in
Figs. 9c,f. These indicate increased levels of heavy and very heavy precipitation at 2 × CO2 . Such
changes are particularly marked in northern India, the Bay of Bengal and the Indochina peninsula,
exceeding changes in the mean. Such increases are statistically significant above the 95% level using
a student t-test (as shown in Figs. 9c,f) and even at the 99% level (not shown). At the 95th percentile
(one in 20 days), events are up to 8mm/day stronger over the northern states (12mm/day over Bay
of Bengal).
Changes to rainfall extremes can also be interpreted by examining the probability of reaching
the absolute level of a given upper percentile at 2 × CO2 . At 1 × CO2 , heavy and very heavy
precipitation events are occurring on 5% and 1% of all days in the season, by definition. Figure 10
measures the increased probability of attaining these absolute levels of precipitation at 2 × CO2 .
Heavy precipitation, as defined at 1 × CO2 , is up to 5% more likely over northern India during JJAS in
HadCM3 2 × CO2 (Fig. 10a). As its previous likelihood was 5%, these events are thus twice as likely.
Similarly, at the very heavy threshold (Fig. 10b), an increase of 2–3% over northern India suggests
very heavy events are more than twice as likely. Hennessy et al. (1997) note a similar divergence in the
response of event return periods to CO2 doubling, i.e., magnified probability increases further up the
distribution. Kharin et al. (2007), however, caution that there are very large uncertainties in changes
to precipitation extremes in the tropics, both amongst models integrated under the SRES scenarios for
12
Figure 11. Spatial pattern of the percentage contribution of subseasonal precipitation extremes to seasonal mean rainfall totals: (a)
contribution of moderately heavy (90th percentile) in HadCM3 1 × CO2 , and (b) the percentage change at 2 × CO2 . The same is
shown for heavy (95th percentile) levels of precipitation in (c) and (d).
IPCC AR4 and observed datasets. Here the spatial changes to absolute levels of the upper extremes or
their probability are strongly tied to the seasonal mean changes shown earlier.
Rather than having entirely negative impacts on society, heavy precipitation, when properly
harnessed, can act as the lifeblood of agriculture and industry, through irrigation systems or reservoirs
storing water for HEP (hydro-electric power generation). To measure the importance of heavy
precipitation events, their contribution to total seasonal rainfall is shown in Fig. 11. This is calculated
on a season-by-season basis before taking the mean, as in Fig. 9, to avoid the effects of interannual
variation. Rainfall above the 90th percentile contributes typically 40% of the seasonal total for much
of India, and this contribution remains as high as 25% even at the 95th percentile. Looking at the
issue of contribution from another angle, Sun et al. (2006) studied the number of days over which
67% of total annual rainfall accumulates. In their study, HadCM3 receives most of its annual total
in less than 30 days, in common with most other coupled GCMs of the IPCC AR4, and broadly
similar to the observed gauge and satellite data compared in their study. At 2 × CO2 (Figs. 11b,d),
contributions from the heaviest events increase particularly over the southeast coast of peninsular India
where summer mean rainfall is low (Fig. 1), indicating increased reliance on heavy precipitation for
the seasonal total to be attained. This will require increased adaptation in the use of water resources
in terms of management, storage, distribution, drainage etc. Together with the reduced probability of
wet days illustrated in Fig. 7(b), this finding indicates a climate more prone to droughts, in agreement
with the modelling study of Semenov and Bengtsson (2002). In contrast to our confined signal, in
a study of trends in the observed record based on Indian station data, Alexander et al. (2006) have
noted positive trends in the contribution of the 95th percentile to the annual total (denoted R95pT)
over most of India for the period 1951–2003. This difference may relate to the data used, or our use
of seasonal instead of annual measures.
Projections of changes to subseasonal extremes in the previous figures are subject to uncertainty
in their spatial distribution, tied to the projected pattern of change in the seasonal mean or the
number of wet days through the season. In order to provide a bulk measure of these changes, and
to summarize these results, Table 1 presents averaged values of mean precipitation intensity, upper
rainfall percentiles and the change in probability of reaching a given percentile. We choose the central
Indian region, after Goswami et al. (2006) of 74.5–86.5◦ E, 16.5–26.5◦ N. As indicated in Table 1 and
Figs. 9–11, mean precipitation intensity generally increases beyond change in the mean, and extremes
13
TABLE 1. Rainfall characteristics for central India (74.5–86.5◦ E, 16.5–26.5◦ N) during JJAS: mean intensity, 90th, 95th and 99th
percentiles (all in mm/day), and the change in probability in reaching the 1 × CO2 levels associated with those percentiles (as
percentage changes).
mean intensity
1 × CO2
2 × CO2
2 × CO2 −1 × CO2
∆prob (1 × CO2 )
(a)
9.7
11.4
1.73
-
extreme percentile
90
95
99
18.3
21.4
3.11
4.3%
24.0
27.9
3.93
3.5%
35.2
40.6
5.92
1.7%
(b)
Figure 12. Levels of precipitation in the upper quartiles of (a) tropics-wide (30◦ N–30◦ S) and (b) all-India rainfall for HadCM3
1 × CO2 (solid) and 2 × CO2 (dashed). Each gridpoint of data is used in each case (no spatial average). Ratio showing their
percentage change is also shown (dotted) and measured on the right-hand axis. After Allen and Ingram (2002).
of precipitation increase even more markedly at 2 × CO2 . The probability of attaining the 1 × CO2
level of precipitation associated with a given extreme also increases at 2 × CO2 in HadCM3 (e.g., the
90th percentiles is 10% likely at 1 × CO2 , but 10 + 4.3 = 14.3% likely at 2 × CO2 ).
Finally, to look at the upper extremes of precipitation in a more detailed manner and determine if
changes to them are in any way predictable, a percentile curve is constructed after Allen and Ingram
(2002). The precipitation data are taken from the upper two quartiles at each gridpoint in the tropicswide region, and at each gridpoint over the Indian land surface (as defined in Section 2.3). Figure 12
shows the level of precipitation at each of the upper percentiles at 1 × CO2 and 2 × CO2 , together
with their ratio. Precipitation increases at 2 × CO2 for all percentiles above ∼ 70% in the tropics-wide
case and over ∼ 60% for India. The response of precipitation associated with given upper percentiles
is relatively flat over India, possibly due to the smaller matrix of input data compared to the tropicswide case; 27 grid-points × 100yr × 120 days, restricting the number of measurable data-points
to ∼ 32 at the 99.99% level. Maximum precipitation increases by 13–14% for both India and the
wider tropics. To investigate the degree to which these increases are in-line with those predicted
using the Clausius-Clapeyron equation (e.g., after Allen and Ingram, 2002, who used 6.5%K−1
based on a global average temperature of 14.2◦ C), the climate sensitivity of the model is calculated.
Defined as the equilibrium response of surface air (1.5m) temperature to CO2 doubling, we limit
the calculation to the tropics (30◦ N–30◦ S), yielding ∆T = +2.27K. The weighted mean temperature
of the same region is 296.3K, yielding via Clausius-Clapeyron an increment rate of 6.05%K−1 and
thus a theoretical maximum precipitation increase of +13.7%. Note that Allen and Ingram (2002)
saw greater increases to the magnitude of the heaviest precipitation in a different configuration of
HadCM3, at around 25%. Reconsideration of the data used in that study revealed that such changes
were really being measured over 2.7 × CO2 , with its corresponding higher climare climate sensitivity
(Pall et al., 2007, their Fig. 1). Allen and Ingram (2002) argue that in the tropics, measured increases
may be more than these predictions due to feedbacks between the release of latent heat and largescale flows supplying moisture. This is particularly true for the India where convective latent heat
release maintains the monsoon following its onset (Sperber et al., 2000). However, the increases in
the heaviest precipitation in HadCM3 do not outpace changes to in-situ moisture availability since
14
there is very little increase in the strength of the broad scale Somali Jet at 2 × CO2 in this model
(Turner et al., 2007). These modelled changes have backing in recent observations. In their study of
the IMD data, Goswami et al. (2006) noted an upward trend of extremes over 1950–2000, especially
at the upper levels where precipitation values which constituted the 99.75 percentile during the 1950s
reached only the 99.5 percentile by the early 1990s. In other modelling studies such as an IS92a
emissions simulation of the 21st century using the ECHAM4/OPYC3 model combination, Semenov
and Bengtsson (2002) used analysis of the Gamma distribution to show positive trends in the scale
parameter, indicating stretching of the probability distribution and exponential increases in heavy
events.
(b)
Changes to active-break cycles
While extreme precipitation events such as those detailed above are undoubtedly important,
consideration needs to be given to both phases of intraseasonal variation as part of active-break cycles
of the Indian summer monsoon. Indeed one of the major unanswered questions of monsoon climate
change research is the impact of increased greenhouse gas forcing on such variations. Their intensity,
frequency of occurrence and duration are vitally important, especially for the sowing and harvesting
of crops.
Composite active and break cycles have been constructed using the all-India rainfall index defined
in section 2. Where the standardised daily rainfall anomaly exceeds one standard deviation (falls below
minus one standard deviation) and persists for at least five days, an active (break) period is defined.
The mid-point of each active or break period during JJAS of each dataset is defined as the lag-zero
reference. Where event duration is even, the mid-point is rounded nearer to the event onset (e.g., if
an event lasts six days then the mid-point is day three). Using this index definition, the mean length
of active (break) periods in HadCM3 is 9.3 days (10.6) versus 7.0 (7.6) using the area-average of the
raw IMD data. Those years in which there are no events of at least five days are excluded from the
mean. In the IMD data, the mean length of events is around 2 days shorter than in HadCM3. This
possibly relates to the tendency for HadCM3 to simulate homogeneous precipitation anomalies over
the whole of India, as suggested in section 3. Greater spatial heterogeneity in the observed data results
in less likelihood of achieving days where all-India rainfall satisfies the active or break threshold.
Model horizontal resolution is the most likely cause of this both directly (e.g., imperfections in the
spatial distribution of convection associated with India’s varied orography) and indirectly (the IMD
and HadCM3 data being on different grid resolutions).
Lag-composites of precipitation anomalies are generated for active and break periods at both
1 × CO2 and 2 × CO2 in HadCM3. These are shown for active event precipitation anomalies to the
annual cycle in HadCM3 1 × CO2 in Fig. 13. Prior to active periods (lag-20), positive anomalies can
be seen building up off the west coast of India, followed by advancement into the south peninsula (and
consistent with Krishnamurthy and Shukla, 2007). These anomalies become a zonal band over much
of central and southern India at lag-12, reaching an amplitude of up to 3mm/day against the seasonal
cycle. Stretching across some 30◦ of longitude, the band becomes more zonally uniform and begins
to expand northward by lag-8. Strong intensification of the anomaly occurs by lag-4, especially west
of Mumbai and in the Bay of Bengal. Around this time, a similar zonal band of negative anomalies
begins to intensify further south, corresponding to reductions in convection over the equatorial TCZ
region, with a minimum around 90◦ E. This represents a dipole in active-break monsoon convection,
between equatorial and Indian latitudes, which is captured in the composites despite the use of
an index based solely on India. Such a dipole is seen more strongly (not shown) when using the
OLR active-break index of Vecchi and Harrison (2002). As time progresses to zero-lag, anomalous
precipitation is maximised over India and the Bay of Bengal, reaching +6–7mm/day. Despite the
chosen index measuring only Indian land-surface rainfall, the composite anomaly peaks over the Bay
of Bengal, illustrating the organized zonal structure of active-break events and greater availability
of moisture over the bay. Modelled patterns of precipitation around the centre of the active period
resemble the quadrupole of convection noted in the observed data by Annamalai and Slingo (2001).
This quadrupole consists of positive anomalies over India / Bay of Bengal and further southeast in the
South China Sea / Indochina, whilst negative anomalies exist over the equatorial Indian Ocean and in
a zonal band across southern China into the East China Sea. As the event progresses past maturity, the
positive precipitation anomalies decline first in the Bay of Bengal and southern states (lag+4), before
moving northwards and shrinking both in magnitude and zonal extent. By lag+12, anomalies largely
disappear.
lag=-24
lag=-20
lag=-16
30N
30N
30N
15N
15N
15N
EQ
EQ
15S
15S
60E
90E
120E
150E
32
EQ
15S
60E
90E
120E
150E
60E
90E
120E
16
lag=-12
lag=-8
lag=-4
30N
30N
30N
15N
15N
15N
EQ
EQ
EQ
15S
15S
60E
90E
120E
150E
8
15S
60E
90E
120E
150E
60E
90E
120E
4
lag=0
lag=+4
lag=+8
30N
30N
30N
15N
15N
15N
EQ
EQ
EQ
15S
15S
60E
90E
120E
150E
15S
60E
lag=+12
90E
120E
150E
60E
lag=+16
30N
30N
15N
15N
15N
EQ
EQ
EQ
15S
15S
120E
150E
120E
1
15S
60E
90E
120E
150E
60E
90E
120E
150E
Lag composites of precipitation anomalies to the seasonal cycle during active events in HadCM3 1 × CO2 , defined by the all-India rainfall index of section 2.3. Units are mm/day, and lag
times are given in days with respect to the centre of an event.
15
Figure 13.
90E
90E
lag=+20
30N
60E
2
16
-1
-4
30N
0.5
-1
-8
0.5
-2
25N
-0.5
-2
20N
16
-0.5
8
15N
4
-4
2
0.5
10N
1
-8
5N
0.5
-0.5
EQ
-1
2
1
-20 -15 -10
-16
-5
0
-4
-8
5
10
15
20
-1
-2
-0.5
Figure 14. Hovmöller composite of anomalous active period vorticity at 850hPa during JJAS, constructed from the active-break
index defined by Fig. 13. Zonally averaged over the Indian region (70 − −90◦ E). Negative anomalies are shaded and have dashed
contours whilst positive anomalies have solid lines. Units of lag (vorticity) are days (10−6 s−1 ).
Compositing over so many events of differing lengths in the 100-year dataset means that
obvious indications of northward propagation can be difficult to see. However, based on the activebreak precipitation index used in Fig. 13, Fig. 14 shows a Hovmöller diagram indicating northward
propagation of lower-tropospheric vorticity anomalies at Indian longitudes. Cyclonic anomalies pass
over the southern tip of peninsular India around 15 days before an active event, intensify, and continue
moving northward until at least 10 days after the event peak. Using the positive cyclonic anomalies in
Fig. 14, we calculate a phase speed of around 0.6ms−1 northwards, consistent with that measured in
the lag-correlations of Fig. 6.
In the break composite (not shown), anomalies evolve in an equal but opposite fashion, although
around zero-lag there are less obvious extensions of the break anomaly to the South China Sea, unlike
the active composites. Monsoon breaks reduce precipitation over India by around 5mm/day during
the event peak (amounting to absolute levels of only 2mm/day), and by almost 9mm/day over the
Bay of Bengal. The quadrupole structure is only visible if the active-break criteria is relaxed to ±0.5σ ,
bringing more events into the composite. Event decay is consistent with anomalies moving to the
north.
To investigate the impact of increased greenhouse gas forcing on active-break cycles of the
monsoon, Fig. 15 shows composites of active and break events at lag-zero. Given that the seasonal
cycle is enhanced at 2 × CO2 (as seen in Fig. 2), absolute levels of precipitation during monsoon
breaks could be broadly similar. Thus we first describe levels of absolute precipitation at 1 × CO2 ,
2 × CO2 and their difference (Fig. 15a,c,e for active and Fig. 15b,d,f for break events). Precipitation
17
(a)
(b)
5
10
7.5
30N
5
5
5
10
5
7.5
5
10
10S
15
7.5
5
5
10
10N
10
5
10S
5
60E
90E
120E
0
150E
60E
10
30N
10
5
5
7.5
10N
7.5
90E
120E
150E
20
7.5
15
(c)
30
(d)
7.5
30N
30N
5
30
10N
10
5
10
5
5
15
5
10N
10
5
5
5
10
10S
10S
5
60E
7.5
5
90E
120E
150E
(e)
60E
90E
120E
150E
(f)
30N
30N
10N
10N
10S
10S
60E
90E
120E
150E
60E
1
0.5
90E
4
2
(g)
120E
150E
16
8
(h)
30N
30N
10N
10N
10S
10S
60E
90E
120E
150E
60E
90E
120E
150E
Figure 15. Absolute levels of precipitation associated with lag-zero composites of active-break cycles in HadCM3: active and
break at 1 × CO2 (a,b); at 2 × CO2 (c,d) and their difference (2 × CO2 minus 1 × CO2 ; e,f). Also shown are the differences
between 2 × CO2 and 1 × CO2 active and break anomalies to the seasonal cycle (g,h). Units are mm/day. In the difference plots,
stippling indicates significance at the 95% level using a student t-test.
18
TABLE 2. Change in precipitation at zero-lag of active and break events averaged over the core monsoon region (Goswami et al.,
2006, 74.5-86.5◦ E, 16.5–26.5◦ N) in response to doubling of CO2 . Differences are shown as an absolute measure (e.g., Figs. 15e,f)
and when changes to the seasonal cycle are taken into account (as in Figs. 15g,h). An asterisk∗ indicates significance beyond the
99% level using a student t-test. Units are mm/day.
precipitation difference
active
break
absolute
anomaly to seasonal cycle
+3.36∗
+2.07∗
+0.12
−1.66∗
during active events is stronger over India at 2 × CO2 , consistent with the arguments presented in
section 4.1. Break events are not measurably more severe in absolute terms at 2 × CO2 (comparing
Figs. 15b,d). By removing the respective daily seasonal cycle of precipitation from each precipitation
dataset using Eq. (1) at each gridpoint, Fig. 15g,h shows the lag-zero composites of the difference
between anomalies to the seasonal cycle at 2 × CO2 and 1 × CO2 . Active and break events are
shown to be more intense against the seasonal cycle, with anomalies increasing in a zonal band across
central India and the Bay of Bengal, and increased anomalies of opposite sign also over the southeast
equatorial Indian Ocean. Over India, these changes are only statistically significant during monsoon
breaks however, indicating that active events at 2 × CO2 are stronger because of the wetter seasonal
cycle, while break events have a larger magnitude against the seasonal cycle. During monsoon breaks,
the region northeast of India which receives additional precipitation (related to northward movement
of the monsoon trough in Krishnamurthy and Shukla, 2007), is increasingly wet.
While there are changes to break event precipitation anomalies for lags of -12 to -8 days (not
shown), these are not statistically significant. From around lag= −6, break events are significantly
drier in the north Bay of Bengal supporting the existing pattern there. Following the event the pattern
of break anomaly intensification persists over the north Bay of Bengal, India, Bangladesh and Nepal
until at least lag= +8.
The above lag-composite analysis has also been carried out based on a precipitation index
measured only over the central core region of India (73–82◦ E, 18–28◦ N), after Mandke et al. (2007).
This yields comparable results to the above in terms of projected increases in the intensity of active
and break event peaks. In addition, changing the minimum length of events used in construction of
the composites (to 3 or 7 days) revealed no major differences.
To summarize changes to precipitation amounts associated with active and break events, we have
taken measurements based on Fig. 15 averaged over the central India region (Goswami et al., 2006)
as in section 4.1. These are displayed in Table 2. In the 2 × CO2 climate of HadCM3, active and
break events are significantly more extreme against the seasonal cycle when compared to 1 × CO2 .
Active events are also significantly wetter in absolute terms. For break events, however, such absolute
changes are statistically insignificant given the increasingly wet mean monsoon. Intensification of
the active-break cycles is caused by overall intensification in the hydrological cycle as the available
moisture content in the atmosphere is increased, as described in section 4.1. Being dipolar in nature,
we speculate that increases in the positive part of the dipole are accompanied by deepening in the
other pole, as part of an intensified local Hadley circulation. Hence break events could be more intense
against the seasonal cycle because of strengthening convection in the eastern equatorial Indian Ocean,
as seen in Figs. 15f,h.
Temporal characteristics of break events such as periodicity and duration could have a dramatic
impact on monsoon-affected societies. However, power spectra of precipitation anomaly timeseries
over India indicate no significant changes to period at 2 × CO2 (not shown). The mean lengths of
active and break events quoted earlier in this section are summarized in Table 3, with the addition of
values for 2 × CO2 . By design, the use of a normalized index will yield a similar number of days
at 1 × CO2 and 2 × CO2 , hence at 2 × CO2 the duration of events relative to the 1 × CO2 event
thresholds are calculated. The table also shows the mean duration of the longest break event in each
season, potentially of more use to society. There are no significant changes to the duration of break
events at 2 × CO2 using either measure. For active events, the mean maximum duration is around 2
days longer at 2 × CO2 , however given the bias of HadCM3 to longer events than in observations,
these difference may be an overestimate. Table 3 shows that re-gridding the observed data to the lower
resolution of HadCM3 accounts for only a small proportion of the model bias.
19
TABLE 3. The mean duration and mean maximum duration of active and break events in HadCM3. Durations at 2 × CO2 are
relative to the 1 × CO2 -defined level of break event precipitation. Also shown are durations for the raw IMD data, and IMD data
interpolated to the model grid (IMDN48 ). Units are days. In the 2 × CO2 column, an asterisk∗ denotes a difference from the
1 × CO2 value statistically significant at the 95% level.
threshold
IMD
IMDN48
1 × CO2
2 × CO2
mean active
mean break
mean max active
mean max break
7.0
7.6
7.1
8.0
7.2
8.3
7.5
8.7
9.3
10.6
10.1
12.2
9.9
12.0
12.4∗
14.0
5. Discussion and conclusions
The greenhouse warming experiment presented here, achieved through the doubling of carbon
dioxide concentrations in the atmosphere of the HadCM3 coupled model, has had several impacts
on the Indian summer monsoon. Aside from an increase in mean precipitation which seems a robust
conclusion among several of the present models, here we have presented future climate changes to
subseasonal monsoon behaviour, drawn from century-long integrations, which fall under three main
headings. Each of the findings outlined is discussed in the context of various biases inherent in global
coupled GCMs.
(a)
Spatial changes to subseasonal monsoon rainfall characteristics
In HadCM3 there is a tendency for a reduction in the number of rain days per season, particularly
in the southeast of India. This occurs in conjunction with increased summer mean rainfall intensity,
which features the same spatial distribution as changes to mean rainfall. The 95th and 99th percentiles
of heavy precipitation on subseasonal timescales are used to measure monsoon extremes and are
found to increase quite strongly, although over India itself these changes are only significant over
the north, where the probability of achieving the 1 × CO2 levels of precipitation is at least doubled.
This suggests a greater potential for damaging floods over the northern region. Changes to extreme
precipitation are more marked over the Bay of Bengal, the Indochina peninsula and much of China,
although for the latter these changes are from a lower baseline. Increases in heavy precipitation are
at the expense of moderate events and are consistent with decreasing number of rain days in some
areas. The potential contribution of extreme rainfall events to the seasonal total has been assessed and
suggests an increased reliance on these events for water supply, particularly in the southeast where
boreal summer rainfall is relatively low and the number of wet days decreases. Such spatial projections
as those outlined above must be interpreted carefully, however. One must question the ability of
coarse resolution GCMs to simulate the spatial nature of extremes of convection in the monsoon,
which featuring large horizontal gradients over regions such as India. In addition, although there is
some agreement on greenhouse warming-related increases in the mean summer monsoon over South
Asia (Meehl et al., 2007), the spatial pattern of such changes on the regional scale is by no means
certain and the results shown here suggest that the spatial pattern of changes to extreme precipitation
are strongly tied to projected changes in the mean. Indeed in a study using the Canadian Centre
for Climate model (CCC GCM2) at 2 × CO2 , Zwiers and Kharin (1998) noted that the strongest
change in precipitation extremes occurred over north-west India, related to intensification of the mean
monsoon there. Despite such limitations, it is important to stress that the spatial distribution of changes
to monsoon extremes depicted here is possible, given the projected mean change found in HadCM3.
The authors are currently examining the relationships between spatial patterns of change in mean
monsoon rainfall and their associated extremes in the models of the CMIP3 database. Once the whole
range of models converge more clearly on the pattern of greenhouse-forced mean monsoon rainfall
change and the mechanisms involved, then findings such as those presented here will be more useful.
We have also shown that even when measuring average changes over a broad region of central India,
large increases in the 95th and 99th percentiles occur.
(b)
Changes to the heaviest monsoon rainfall
To assess the impact of increased greenhouse gas forcing on rainfall extremes over India as a
whole, increases in the magnitude of the heaviest precipitation events were measured. Looking at all
model data over India and tropics-wide, the increased magnitude of the heaviest rainfall is around
20
+13–14%, in broad agreement with the predicted increase in the water holding capacity of the atmospheric column and the level of surface warming experienced in this model. Thus projected increases
in the heaviest rainfall over India seem remarkably predictable. However, models such as HadCM3
are unable to simulate the upper tail of precipitation extremes (Fig. 3b) owing to their resolution. In
higher resolution and/or regional climate models, the consequent improved representation of localized
deep convection leads to a more realistic (increased) frequency of extreme events (Jones et al., 1997).
Consequently, they argue that in higher resolution models, smaller fractional increases are measured in
the frequency of upper extremes of precipitation as CO2 concentrations are increased, together with
corresponding larger increases in their magnitude. However the results presented here suggest that
HadCM3 has already reached the thermodynamic limit predicted by the Clausius-Clapeyron relation,
suggesting that such changes may not be restricted by grid size and are independent of any bias in
the seasonal cycle. HadCM3 also lacks a significant dynamic contribution to increase in extremes
of monsoon rainfall, given little climate-induced changes to the Somali Jet. Work is in progress to
examine if this thermodynamic component is predictable in other coupled GCMs and to elucidate
the relative role of any dynamical component. In models which show increases in seasonal mean
monsoon rainfall, the Clausius-Clapeyron relation may therefore provide a lower bound on increases
in the heaviest rainfall over monsoon regions, supporting the view held by Allen and Ingram (2002).
(c)
Changes to active-break cycles
Break events have been shown to be more intense against the annual cycle in HadCM3 at
2 × CO2 , in common with other models (e.g., Mandke et al., 2007). Although these could have
important repercussions, particularly for agriculture (heat stress of crops etc.), the fact that their
duration and periodicity remain unchanged suggests this may not cause serious problems. More
noticeable is the intensification of active events (extended periods of heavy rainfall), which could
lead to a greater potential for crop and infrastructure damage as well as loss of life from flooding.
The results above are still clear when measured as an average over the central India region. Clearly
the ability of GCMs to simulate the intricacies of monsoon intraseasonal oscillations, particularly
any biases in their propagation characteristics, is important. We have demonstrated that HadCM3
does have some skill in simulating northward propagation of ISO in the Indian monsoon domain,
comparable with other GCMs (Lin et al., 2008). This propagation is still deficient when compared
against observed precipitation, likely related to weak air-sea coupled interactions in the model. The
representation of such processes could potentially be improved by increasing the atmosphere-ocean
coupling frequency and improving the vertical resolution of the upper ocean (Bernie et al., 2005).
That HadCM3 at least features some characteristics inherent to monsoon ISO, as well as its mean
and seasonal cycle being among the best of the current models (Annamalai et al., 2007), and the
large sample size from which these results are drawn, suggests we may have some confidence in
the projections of active-break cycles made here. The reasons for the intensification of break events
remain unclear, but we speculate that they relate to intensification in the local Hadley circulation, and
they remain the subject of a future study.
This paper thus provides a projection of changes to intraseasonal behaviour in the Asian summer
monsoon, comprising generally wetter extremes, the upper limit of which seems quite predictable
based on the degree of projected warming. Active and break events were found to become more
intense against the seasonal cycle. The array of metrics presented will be useful for assessing future,
higher resolution GCMs, featuring more realistic ocean-atmosphere coupled processes.
Acknowledgements
A. G. Turner is funded via the EU-ENSEMBLES project, while Julia Slingo is a member of the
NERC National Centre for Atmospheric Science (NCAS). The authors are grateful to two anonymous
reviewers and the associate editor, Oscar Alves, for their constructive criticism of this work. The India
Meteorological Department’s gridded rainfall dataset was made available by M. Rajeevan, CMAP was
obtained from the Climate Prediction Center, and GPCP data was obtained from the NASA Goddard
Space Flight Center. Computing resources were provided by HPCx for running the Unified Model.
21
References
Adler RF, Huffman GJ, Chang A, Ferraro R, Xie P, Janowiak J, Rudolf B, Schneider U, Curtis S, Bolvin D, Gruber A, Susskind J, Arkin P.
2003. The version 2 global precipitation climatology project (gpcp) monthly precipitation analysis (1979-present). J. Hydrometeorol. 4:
1147–1167. DOI: 10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2.
Alexander LV, Zhang X, Peterson TC, Caesar J, Gleason B, Klein Tank AMG, Haylock M, Collins D, Trewin B, Rahimzadeh F, Tagipour A,
Rupa Kumar K, Revadekar J, Griffiths G, Vincent L, Stephenson DB, Burn J, Aguilar E, Brunet M, Taylor M, New M, Zhai P, Rusticucci
M, Vasquez-Aguirre JL. 2006. Global observed changes in daily climate extremes of temperature and precipitation. J. Geophys. Res.
111. DOI:10.1029/2005JD006290.
Allen MR, Ingram WJ. 2002. Constraints on future changes in climate and the hydrologic cycle. Nature 419: 224–232.
DOI:10.1038/nature01092.
Annamalai H, Hamilton K, Sperber KR. 2007. The South Asian Summer Monsoon and Its Relationship with ENSO in the IPCC AR4
Simulations. J. Climate 20: 1071–1092. DOI: 10.1175/JCLI4035.1.
Annamalai H, Slingo JM. 2001. Active/break cycles: diagnosis of the intraseasonal variability of the Asian Summer Monsoon. Clim. Dyn.
18: 85–102. DOI:10.1007/s003820100161.
Ashrit R, Douville H, Kumar K. 2003. Response of the Indian Monsoon and ENSO-Monsoon Teleconnection to Enhanced Greenhouse
Effect in the CNRM Coupled Model. J. Meteorol. Soc. Jpn. 81: 779–803. DOI: 10.2151/jmsj.81.779.
Bernie DJ, Woolnough SW, Slingo JM, Guilyardi E. 2005. Modelling diurnal and intraseasonal variability of the ocean mixed layer. J.
Climate 18: 1190–1202. DOI: 10.1175/JCLI3319.1.
Dai A, Trenberth KE. 2004. The Diurnal Cycle and Its Depiction in the Community Climate System Model. J. Climate 17: 930–951. DOI:
10.1175/1520-0442(2004)017<0930:TDCAID>2.0.CO;2.
Duchon CE. 1979. Lanczos Filtering in One and Two Dimensions. Journal of Applied Meteorology 18: 1016–1022. DOI: 10.1175/15200450(1979)018<1016:LFIOAT>2.0.CO;2.
Fu X, Wang B, Li T, McCreary JP. 2003. Coupling between Northward-Propagating, Intraseasonal Oscillations and Sea Surface Temperature
in the Indian Ocean. J. Atmos. Sci. 60: 1733–1271. DOI: 10.1175/1520-0469(2003)060<1733:CBNIOA>2.0.CO;2.
Gadgil S, Sajani S. 1998. Monsoon precipitation in the AMIP runs. Clim. Dyn. 14: 659–689. DOI: 10.1007/s003820050248.
Gordon C, Cooper C, Senior CA, Banks H, Gregory JM, Johns TC, Mitchell JFB, Wood RA. 2000. The simulation of SST, sea ice extents
and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Clim. Dyn. 16: 147–168. DOI:
10.1007/s003820050010.
Goswami BN, Mohan RSA. 2001. Intraseasonal Oscillations and Interannual Variability of the Indian Summer Monsoon. J. Climate 14:
1180–1198. DOI: 10.1175/1520-0442(2001)014%3C1180:IOAIVO%3E2.0.CO;2.
Goswami BN, Venugopal V, Sengupta D, Madhusoodanan MS, Xavier PK. 2006. Increasing Trend of Extreme Rain Events over India in a
Warming Environment. Science 314: 1442–1445. DOI: 10.1126/science.1132027.
Goswami BN, Xavier PK. 2005. Dynamics of ”internal” interannual variability of the Indian summer monsoon in a GCM. J. Geophys. Res.
110. DOI: 10.1029/2005JD006042.
Hennessy KJ, Greogry JM, Mitchell JFB. 1997. Changes in daily precipitation under enhanced greenhouse conditions. Clim. Dyn. 13:
667–680. DOI: 10.1007/s003820050189.
Hu ZZ, Latif M, Roeckner E, Bengtsson L. 2000. Intensified Asian summer monsoon and its variability in a coupled model forced by
increasing greenhouse gas concentrations. Geophys. Res. Lett. 27: 2618–2684. DOI: 10.1029/2000GL011550.
Huffman G, Adler RF, Morrissey M, Bolvin DT, Curtis S, Joyce R, McGavock B, Susskind J. 2001. Global precipitation at one-degree daily
resolution from multi-satellite observations. J. Hydrometeor. 2: 36–50. DOI: 10.1175/1525-7541(2001)002¡0036:GPAODD¿2.0.CO;2.
Inness P, Slingo J, Woolnough S, Neale R, Pope V. 2001. Organization of tropical convection in a GCM with varying vertical resolution:
implications for the simulation of the Madden-Julian Oscillation. Clim. Dyn. 17: 777–793. DOI: 10.1007/s003820000148.
Johns TC, Gregory JM, Ingram WJ, Johnson CE, Jones A, Lowe JA, Mitchell JFB, Roberts DL, Sexton DMH, Stevenson DS, Tett SFB,
Woodage MJ. 2003. Anthropogenic climate change for 1860 to 2100 simulated with the HadCM3 model under updated emissions
scenarios. Clim. Dyn. 20: 583–612. DOI: 10.1007/s00382-002-0296-y.
Jones RG, Murphy JM, Noguer M, Keen AB. 1997. Simulation of climate change over Europe using a nested regional-climate model. II:
Comparison of driving and regional modelling responses to a doubling of carbon dioxide. Q. J. R. Meteorol. Soc. 123: 265–292. DOI:
10.1002/qj.49712353802.
Joseph PV, Simon A. 2005. Weakening trend of the southwest monsoon current through peninsular India from 1950 to the present. Current
Science 89: 687–694.
Kemball-Cook S, Wang B. 2001. Equatorial Waves and Air-Sea Interaction in the Boreal Summer Intraseasonal Oscillation. J. Climate 14:
2923–2942. DOI: 10.1175/1520-0442(2001)014<2923:EWAASI>2.0.CO;2.
Kharin VV, Zwiers FW, Zhang X, Hegerl GC. 2007. Changes in Temperature and Precipitation Extremes in the IPCC Ensemble of Global
Coupled Model Simulations. J. Climate 20: 1419–1444. DOI: 10.1175/JCLI4066.1.
Kitoh A, Yukimoto S, Noda A, Motoi T. 1997. Simulated Changes in the Asian Summer Monsoon at Times of Increased Atmospheric CO2 .
J. Meteorol. Soc. Jpn. 75: 1019–1031.
Krishnamurthy V, Shukla J. 2000. Intraseasonal and Interannual Variability of Rainfall over India. J. Climate 13: 4366–4377. DOI:
10.1175/1520-0442(2000)013<0001:IAIVOR>2.0.CO;2.
Krishnamurthy V, Shukla J. 2007. Intraseasonal and Seasonally Persisting Patterns of Indian Monsoon Rainfall. J. Climate 20: 3–20. DOI:
10.1175/JCLI3981.1.
Krishnamurthy V, Shukla J. 2008. Intraseasonal and Seasonally Persisting Patterns of Indian Monsoon Rainfall. Clim. Dyn. 30: 353–369.
DOI: 10.1007/s00382-007-0300-7.
Lal B, Jayanthi N, abd M Rajeevan TP, Devi S, Srivastava AK, Pai DS. 2006. Monsoon 2005 : A report. National Climate Centre, Indian
Meteorological Department, Pune.
22
Lin JL, Weickmann KM, Kiladis GN, Mapes BE, Schubert SD, Suarez MJ, Bacmeister JT, Lee MI. 2008. Subseasonal Variability Associated
with Asian Summer Monsoon Simulated by 14 IPCC AR4 Coupled GCMs. J. Climate 21: 4541–4567. DOI: 10.1175/2008JCLI1816.1.
Mandke SK, Sahai AK, Shinde MA, Joseph S, Chattopadhyay R. 2007. Simulated changes in active/break spells during the Indian
summer monsoon due to enhanced CO2 concentrations: assessment from selected coupled atmosphere-ocean global climate models.
International Journal of Climatology 27: 837–859. DOI: 10.1002/joc.1440.
Martin GM, Arpe K, Chauvin F, Ferranti L, Maynard K, Polcher J, Stephenson DB, Tschuck P. 2000. Simulation of the Asian Summer
Monsoon in Five European General Circulation Models. Atmos. Sci. Let. 1: 37–55. DOI: 10.1006/asle.2000.0004.
May W. 2004. Potential future changes in the Indian summer monsoon due to greenhouse warming: analysis of mechanisms in a global
time-slice experiment. Clim. Dyn. 22: 389–414. DOI: 10.1007/s00382-003-0389-2.
Meehl G, Collins WD, Boville BA, Kiehl JT, Wigley TML, Arblaster JM. 2000. Response of the NCAR climate system model to increased
CO2 and the role of physical processes. J. Climate 13: 1879–1898. DOI: 10.1175/1520-0442(2000)013<1879:ROTNCS>2.0.CO;2.
Meehl G, Stocker T, Collins W, Friedlingstein P, Gaye A, Gregory J, Kitoh A, Knutti R, Murphy J, Noda A, Raper S, Watterson I, Weaver A,
Zhao ZC. 2007. Global Climate Projections, ch. 10. Cambridge University Press, pp. 747–846. In: Climate Change 2007: The Physical
Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.
Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor, H.L. Miller (eds.).
Meehl GA, Arblaster JM. 2003. Mechanisms for projected future changes in south Asian monsoon precipitation. Clim. Dyn. 21: 659–675.
DOI: 10.1007/s00382-003-0343-3.
Pall P, Allen MR, Stone DA. 2007. Testing the Clausius-Clapeyron constraint on changes in extreme precipitation under CO2 warming.
Clim. Dyn. 28: 351–363. DOI: 10.1007/s00382-006-0180-2.
Perkins SE, Pitman AJ, Holbroock NJ, McAneney J. 2007. Evaluation of the AR4 Climate Models’ Simulated Daily Maximum Temperature,
Minimum Temperature, and Precipitation over Australia Using Probability Density Functions. J. Climate 20: 4356–4376. DOI:
10.1175/JCLI4253.1.
Pope VD, Gallani ML, Rowntree PR, Stratton RA. 2000. The impact of new physical parametrizations in the Hadley Centre climate model:
HadAM3. Clim. Dyn. 16: 123–146. DOI: 10.1007/s003820050009.
Rajeevan M, Bhate J, Kale JD, Lal B. 2006. High resolution daily gridded rainfall data for the Indian region: Analysis of break and active
monsoon spells. Current Science 91: 296–306.
Semenov VA, Bengtsson L. 2002. Secular trends in daily precipitation charactersitics: greenhouse gas simulation with a coupled AOGCM.
Clim. Dyn. 19: 123–140. DOI: 10.1007/s00382-001-0218-4.
Sen Roy S, Balling Jr RC. 2004. Trends in extreme daily precipitation indices in India. International Journal of Climatology 24: 457–466.
DOI: 10.1002/joc.995.
Spencer H, Slingo JM. 2003. The simulation of peak and delayed ENSO teleconnections. J. Climate 16: 1757–1774. DOI: 10.1175/15200442(2003)016<1757:TSOPAD>2.0.CO;2.
Sperber KR, Slingo JM, Annamalai H. 2000. Predictability and the relationship between subseasonal and interannual variability during the
Asian summer monsoon. Q. J. R. Meteorol. Soc. 126: 2545–2574. DOI: 10.1002/qj.49712656810.
Stephenson DB, Kumar KR, Doblas-Reyes FJ, Royer JF, Chauvin F. 1999. Extreme Daily Rainfall Events and Their Impact on Ensemble
Forecasts of the Indian Monsoon. Mon. Weather Rev. 127: 1954–1966. DOI: 10.1175/1520-0493(1999)127<1954:EDREAT>2.0.CO;2.
Sun Y, Solomon S, Dai A, Portman RW. 2006. How often does it rain? J. Climate 19: 916–934. DOI: 10.1175/JCLI3672.1.
Sun Y, Solomon S, Dai A, Portman RW. 2007. How often will it rain? J. Climate 20: 4801–4818. DOI: 10.1175/JCLI4263.1.
Trenberth KE, Dai A, Rasmussen RM, Parsons DB. 2003. The changing character of precipitation. Bull. Am. Meteorol. Soc. September:
1205–1217. DOI: 10.1175/BAMS-84-9-1205.
Turner AG, Inness PM, Slingo JM. 2005. The role of the basic state in the ENSO-monsoon relationship and implications for predictability.
Q. J. R. Meteorol. Soc. 131: 781–804. DOI: 10.1256/qj.04.70.
Turner AG, Inness PM, Slingo JM. 2007. The effect of doubled CO2 and model basic state biases on the monsoon-ENSO system. I: Mean
response and interannual variability. Q. J. R. Meteorol. Soc. 133: 1143–1157. DOI: 10.1002/qj.82.
Vecchi GA, Harrison DE. 2002. Monsoon Breaks and Subseasonal Sea Surface Temperature Variability in the Bay of Bengal. J. Climate
15: 1485–1493. DOI: 10.1175/1520-0442(2002)015¡1485:MBASSS¿2.0.CO;2.
Webster PJ, Magana VO, Palmer TM, Shukla J, Tomas RA, Yanai M, Yasunari T. 1998. Monsoons: Processes, predictability, and prospects
for predcition. J. Geophys. Res. 103: 14 451–14 510. DOI: 10.1029/97JC02719.
Xie P, Arkin PA. 1997. Global Precipitation: A 17-Year Monthly Analysis Based on Gauge Observations, Satellite Estimates, and Numerical
Model Outputs. Bull. Am. Meteorol. Soc. 78: 2539–2559. 10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2.
Zwiers FW, Kharin VV. 1998. Changes in the Extremes of the Climate Simulated by CCC GCM2 under CO2 Doubling. J. Climate 11:
2200–2222. DOI: 10.1175/1520-0442(1998)011<2200:CITEOT>2.0.CO;2.